Web Comm, 2017-18 edition

Minutes for lecture IV

Searching for a (good) solution

Main classes of problems:
	optimizaton, clusterization, classification (, ordering)
	
	-knapsac (optimization)
	-travel planning (optimization)
	-voice/face recognition (classification)
	-work best match: soundness, ortographic correction (classification)
	-web search engines (clustering, ordering)
	
All is fitting functions
	given a set of possible solutions find the best solution 
	that maximize/mnimize a "cost" function.
	
Optimal solution vs aproximation
	for all 'interesting' problem its "hard" to find best solution
	we have to find something similar (near) to the best: 
	we need a search path

gradient method (greedy)
	move myself toward the best I can see
	When I have stop myself? 
	The local-maxima problem.
	
	two examples on matlab

Sometime randomness is useful
	The Montecarlo method for calculates Pi greek.
	
	an example on matlab
	
ant method
	move myself randomly around for a while and move myself to the best I found 

genetic method
	take a set of candidates, randomly combine it in order to find other candidates,
	select a new set of candidates from it